AI-Native
Communications.
An internal exploration into what calling, messaging, and presence look like when AI is at the core of communication, not a feature added on top. Two designers, the full comms surface, intentionally greenfield.
Setting up context.
The project. An internal exploration into what communications could look like when AI is at the core: what happens to calling, messaging, and presence in a world where AI is native, not added on.
My scope. Just two designers, covering the entire comms surface: messaging, calling, contacts, presence. No brief beyond the core question. Since this was an exploration sprint, all constraints were intentionally lifted.
The brief was the problem.
Before designing anything, I had to understand what communication fundamentally is underneath the app layer. What is a phone call? What job is a user trying to do, before apps were the answer to that job?
Concepts, not screens.
Deconstructed the fundamental intent
Of calling and messaging: what they exist to do for users, stripped of any product layer.
Mapped the behavioral shift required
Moving from “open an app” to a free-flowing conversation with an AI. UX must shift from navigating interfaces to declaring intent.
Made the primary artifact a Google Doc
Working in screens too early anchors you to the wrong level of abstraction.
Four levels of AI participation.
Direct Connection
AI just opens the pipe and gets out of the way. It acts only as a voice-activated dialer. AI’s role in communication today.
Background Assist
The user is in the conversation, AI is passively monitoring and surfacing context.
Negotiated Handoff
AI handles the setup, screening, and coordination, then hands off to the user for the live conversation.
Full Delegation
AI handles the entire interaction end-to-end, user is never involved live. Notifies user after task has completed.
Full Delegation isn’t one thing.
Agent × Human
The agent interacts with a person on the other end. It must navigate tone, social etiquette, emotional cues, reading the room: all the unwritten contracts of human communication.
Agent × Agent
The agent interacts with another AI independently. It exchanges information with each other based on the social guidance of their persons. How can we start to define the rules around agent to agent interactions?
Ended the sprint with a prototype.
The sprint ended with a prototype: a high-level look at what agents do in the background on a user’s behalf, and how that gets surfaced back to them.
The hard questions still open.
Relationships are complex
To get this right, agents need to map the nuances of how people relate to one another: not just who you’re talking to, but how. That’s a much harder modeling problem than a contact list.
Social intelligence
Etiquette, emotional cues, reading the room. Are our agents intelligent enough to factor these in? And if they’re not, we need to design around that gap, not past it.
Safety and privacy
For any of this to work at scale, users have to trust the agent isn’t doing more than they sanctioned. That’s not a legal checkbox; it’s a design challenge.
The social profile problem
A profile may not be a simple list. To truly capture someone’s social comfort, it may need to be far more complex and dynamic than anything we’ve designed before.
What does it actually mean to represent someone’s relationships faithfully enough for an agent to act on their behalf?
Vocabulary and direction.
This didn’t ship
That’s the honest answer. This was an exploration. The value was never going to be a launched product, and I want to name that directly.
The framework gave the org a shared vocabulary
The four levels of AI participation gave teams a way to talk about AI’s role in communication that wasn’t just “add AI features.” It named the design space.
Agent-to-agent seeded a real workstream
The distinction got picked up and became an active prototype investment area. The exploration gave the org a direction it could act on, which is what good exploratory design should produce.